Skip to contentSkip to conversation

KVM: arm64: Fix 64 bit mmio handle

PATCH In review Schspa Shi 2 participants 9 messages Jun 30, 2022 – Jul 6, 2022

Subsystems (2)
lore
Share & tools
lore

Schspa Shi

author
#0 lore raw

PATCH 1/1 KVM: arm64: Fix 64 bit mmio handle

If the len is 8 bytes, we can't get the correct sign extend for be system.

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +2 -2
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
 
 	/* Detect an already handled MMIO return */
 	if (unlikely(!vcpu->mmio_needed))
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 		data = kvm_mmio_read_buf(run->mmio.data, len);
 
 		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {
 			mask = 1U << ((len * 8) - 1);
 			data = (data ^ mask) - mask;
 		}
Signature · 1 line
2.37.0
maintainer
#1 lore raw
Quoted 3 lines · Schspa Shi

If the len is 8 bytes, we can't get the correct sign extend for be system.

I'm afraid you'll have to give me a bit more details.

Quoted 23 lines

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +1 -1
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
 
 	/* Detect an already handled MMIO return */
 	if (unlikely(!vcpu->mmio_needed))

arch/arm64/kvm/mmio.c int kvm_handle_mmio_return(struct kvm_vcpu *vcpu) · line 100

Quoted 5 lines
arch/arm64/kvm/mmio.c +1 -1
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 		data = kvm_mmio_read_buf(run->mmio.data, len);
 
 		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {

If you're reading an 8 byte quantity, what is there to sign-extend? Sign extension only makes sense if what you're reading is *smaller* than the size of the register you are targeting.

I must be missing something. And how is that related to running BE? BE in the host? The guest?

Please convince me.

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
author
#2 lore raw
Quoted 35 lines · Marc Zyngier
Quoted 3 lines · Schspa Shi

If the len is 8 bytes, we can't get the correct sign extend for be system.

I'm afraid you'll have to give me a bit more details.

Quoted 27 lines

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +1 -1
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void 
 *buf, unsigned int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
 
 	/* Detect an already handled MMIO return */
arch/arm64/kvm/mmio.c +0 -0
 	if (unlikely(!vcpu->mmio_needed))
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu 
 *vcpu)
 		data = kvm_mmio_read_buf(run->mmio.data, len);
 
Quoted 8 lines

arch/arm64/kvm/mmio.c int kvm_handle_mmio_return(struct kvm_vcpu *vcpu) · line 100

Quoted 3 lines
arch/arm64/kvm/mmio.c +1 -1
 		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {

If you're reading an 8 byte quantity, what is there to sign-extend? Sign extension only makes sense if what you're reading is *smaller* than the size of the register you are targeting.

Yes, you are correct, sorry for my bad patch.
Please ignore this patch.

Quoted 2 lines

I must be missing something. And how is that related to running BE? BE in the host? The guest?

I mean BE is for guest running with BE mode.

Quoted 4 lines

Please convince me.

	M.
Signature · 2 lines
BRs
Schspa Shi
maintainer
#3 lore raw
Quoted 43 lines · Schspa Shi
Quoted 31 lines · Marc Zyngier
Quoted 3 lines · Schspa Shi

If the len is 8 bytes, we can't get the correct sign extend for be system.

I'm afraid you'll have to give me a bit more details.

Quoted 23 lines

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +1 -1
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void *buf, 
 unsigned int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
  	/* Detect an already handled MMIO return */
 	if (unlikely(!vcpu->mmio_needed))
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
Quoted 9 lines

arch/arm64/kvm/mmio.c int kvm_handle_mmio_return(struct kvm_vcpu *vcpu) · line 100

Quoted 4 lines
arch/arm64/kvm/mmio.c +1 -1
 		data = kvm_mmio_read_buf(run->mmio.data, len);
  		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {

If you're reading an 8 byte quantity, what is there to sign-extend? Sign extension only makes sense if what you're reading is *smaller* than the size of the register you are targeting.

Quoted 7 lines

Yes, you are correct, sorry for my bad patch.
Please ignore this patch.

Quoted 2 lines

I must be missing something. And how is that related to running BE? BE in the host? The guest?

I mean BE is for guest running with BE mode.

So what problem did you see? If you have noticed something going wrong, I'd like to get it fixed.

Thanks,

         M.
Signature · 1 line
Jazz is not dead. It just smells funny...
author
#4 lore raw
Quoted 42 lines · Marc Zyngier
Quoted 41 lines · Schspa Shi
Quoted 27 lines · Marc Zyngier
Quoted 2 lines · Schspa Shi

If the len is 8 bytes, we can't get the correct sign extend for be system.

I'm afraid you'll have to give me a bit more details.

Quoted 21 lines

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +1 -1
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void 
 *buf, unsigned
 int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
  	/* Detect an already handled MMIO return */
arch/arm64/kvm/mmio.c +0 -0
 	if (unlikely(!vcpu->mmio_needed))
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu 
Quoted 9 lines

arch/arm64/kvm/mmio.c int kvm_handle_mmio_return(struct kvm_vcpu *vcpu) · line 100

Quoted 5 lines
arch/arm64/kvm/mmio.c +1 -1
 *vcpu)
 		data = kvm_mmio_read_buf(run->mmio.data, len);
  		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {

If you're reading an 8 byte quantity, what is there to sign-extend? Sign extension only makes sense if what you're reading is *smaller* than the size of the register you are targeting.

Yes, you are correct, sorry for my bad patch.
Please ignore this patch.

Quoted 7 lines
Quoted 3 lines
Quoted 2 lines

I must be missing something. And how is that related to running BE? BE in the host? The guest?

I mean BE is for guest running with BE mode.

So what problem did you see? If you have noticed something going wrong, I'd like to get it fixed.

I have running some static code analysis software upon Kernel code. Seeing there is possible overflow.

maks << 1U << ((len * 8) -1);

The AI don't know, len is only the value of 1, 2, 4, and make this a warnings

I tring to analysis this, but didn't realize the real scenario of sign extension, and finally sent this problematic patch.

I do see some uninitialized memory reads (the values are not used in the end, just as temporary space for API execution), do we need to fix these?

Quoted 3 lines

Thanks,

         M.
Signature · 2 lines
Schspa Shi
BRs
maintainer
#5 lore raw
Quoted 65 lines · Schspa Shi
Quoted 44 lines · Marc Zyngier
Quoted 43 lines · Schspa Shi
Quoted 29 lines · Marc Zyngier
Quoted 2 lines · Schspa Shi

If the len is 8 bytes, we can't get the correct sign extend for be system.

I'm afraid you'll have to give me a bit more details.

Quoted 23 lines

Fix the mask type len and the comparison of length.

---
 arch/arm64/kvm/mmio.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/arm64/kvm/mmio.c +1 -1
index 3dd38a151d2a6..0692f8b18f35c 100644
@@ -81,8 +81,8 @@ unsigned long kvm_mmio_read_buf(const void
 *buf, unsigned
 int len)
 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 {
 	unsigned long data;
+	unsigned long mask;
 	unsigned int len;
-	int mask;
  	/* Detect an already handled MMIO return */
arch/arm64/kvm/mmio.c +0 -0
 	if (unlikely(!vcpu->mmio_needed))
@@ -97,7 +97,7 @@ int kvm_handle_mmio_return(struct kvm_vcpu
 *vcpu)
 		data = kvm_mmio_read_buf(run->mmio.data, len);
Quoted 9 lines

arch/arm64/kvm/mmio.c int kvm_handle_mmio_return(struct kvm_vcpu *vcpu) · line 100

Quoted 3 lines
arch/arm64/kvm/mmio.c +1 -1
  		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
+		    len <= sizeof(unsigned long)) {

If you're reading an 8 byte quantity, what is there to sign-extend?
Sign extension only makes sense if what you're reading is *smaller*
than the size of the register you are targeting.

Yes, you are correct, sorry for my bad patch.
Please ignore this patch.

Quoted 8 lines
Quoted 4 lines
Quoted 3 lines

I must be missing something. And how is that related to running BE? BE
in the host? The guest?

I mean BE is for guest running with BE mode.

So what problem did you see? If you have noticed something going wrong, I'd like to get it fixed.

I have running some static code analysis software upon Kernel code. Seeing there is possible overflow.

maks << 1U << ((len * 8) -1);

The AI don't know, len is only the value of 1, 2, 4, and make this a warnings

Quoted 6 lines

I tring to analysis this, but didn't realize the real scenario of sign extension, and finally sent this problematic patch.

I do see some uninitialized memory reads (the values are not used in the end, just as temporary space for API execution), do we need to fix these?

You need to be more descriptive here. What uninitialised reads? In general, pointing at the code and providing a full description of what you think is incorrect would really help...

	M.
Signature · 5 lines
Without deviation from the norm, progress is not possible.
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author
#6 lore raw
Quoted 16 lines · Marc Zyngier
Quoted 9 lines

I have running some static code analysis software upon Kernel code. Seeing there is possible overflow.

maks << 1U << ((len * 8) -1);

The AI don't know, len is only the value of 1, 2, 4, and make this a warnings

Quoted 6 lines

I tring to analysis this, but didn't realize the real scenario of sign extension, and finally sent this problematic patch.

I do see some uninitialized memory reads (the values are not used in the end, just as temporary space for API execution), do we need to fix these?

Quoted 5 lines

You need to be more descriptive here. What uninitialised reads? In general, pointing at the code and providing a full description of what you think is incorrect would really help...

	M.

One example is
int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)

{
	...
   case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
		u64 reg, id;

		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
	}

}

The funcion vgic_v3_has_cpu_sysregs_attr will read reg's value to params without initialization. There should have no problems, because the register value never used.

Signature · 7 lines
BRs
Schspa Shi

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
maintainer
#7 lore raw
Quoted 38 lines · Schspa Shi
Quoted 16 lines · Marc Zyngier
Quoted 9 lines

I have running some static code analysis software upon Kernel code. Seeing there is possible overflow.

maks << 1U << ((len * 8) -1);

The AI don't know, len is only the value of 1, 2, 4, and make this a warnings

Quoted 6 lines

I tring to analysis this, but didn't realize the real scenario of sign extension, and finally sent this problematic patch.

I do see some uninitialized memory reads (the values are not used in the end, just as temporary space for API execution), do we need to fix these?

Quoted 6 lines

You need to be more descriptive here. What uninitialised reads? In general, pointing at the code and providing a full description of what
you think is incorrect would really help...

	M.

One example is
int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)

{
	...
   case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
		u64 reg, id;

		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
	}
Quoted 5 lines

}

The funcion vgic_v3_has_cpu_sysregs_attr will read reg's value to params without initialization. There should have no problems, because the register value never used.

Thanks for pointing this out.

I spent some time looking at this, and this is only the tip of the iceberg. The whole userspace interaction with the GIC sysregs is ugly (at best), and needs some love.

I've written a small series[1] cleaning things up, which needs testing (I've just checked that it was compiling correctly). I'd appreciate you running your tool on it.

	M.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/log/?h=kvm-arm64/sysreg-cleanup-5.20

Signature · 5 lines
Without deviation from the norm, progress is not possible.
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author
#8 lore raw
Quoted 48 lines · Marc Zyngier
Quoted 38 lines · Schspa Shi
Quoted 16 lines · Marc Zyngier
Quoted 9 lines

I have running some static code analysis software upon Kernel code. Seeing there is possible overflow.

maks << 1U << ((len * 8) -1);

The AI don't know, len is only the value of 1, 2, 4, and make this a warnings

Quoted 6 lines

I tring to analysis this, but didn't realize the real scenario of sign extension, and finally sent this problematic patch.

I do see some uninitialized memory reads (the values are not used in the end, just as temporary space for API execution), do we need to fix these?

Quoted 6 lines

You need to be more descriptive here. What uninitialised reads? In general, pointing at the code and providing a full description of what
you think is incorrect would really help...

	M.

One example is
int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)

{
	...
   case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
		u64 reg, id;

		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
	}
Quoted 5 lines

}

The funcion vgic_v3_has_cpu_sysregs_attr will read reg's value to params without initialization. There should have no problems, because the register value never used.

Thanks for pointing this out.

Quoted 8 lines

I spent some time looking at this, and this is only the tip of the iceberg. The whole userspace interaction with the GIC sysregs is ugly (at best), and needs some love.

I've written a small series[1] cleaning things up, which needs testing (I've just checked that it was compiling correctly). I'd appreciate you running your tool on it.

I have run static code analysis software upon this branch, and the warnings have gone.

Quoted 3 lines
	M.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/log/?h=kvm-arm64/sysreg-cleanup-5.20

Signature · 2 lines
BRs
Schspa Shi